Fuel–food nexus in urban areas: evidence from Burkina Faso

ABSTRACT

This study examines the transmission of fuel prices to food security among households with motorcycles in Ouagadougou, Burkina Faso, combining quantitative and behavioural analyses. The results indicate that approximately 61.3% of households were affected by food insecurity between 2018 and 2019. This share comprises those experiencing meagre forms of food insecurity (24.8%), moderate food insecurity (28.3%), and the most severely affected (8.2%). One of the chief reasons for food insecurity is households’ high reliance on motorcycles as a primary means of transportation. Low-income levels and unproductive rides can reinforce exposure to such vulnerability. Besides, households react differently and asymmetrically to fuel price changes. Reactions to hypothetical fluctuations in fuel prices suggest a positive association between gradual increases in fuel prices and food insecurity. Households’ exposure to food insecurity is further bolstered when the head is a female, non-salaried, less educated, of low income, or from a large household.

KEYWORDS:

• Food security
• energy prices
• Burkina Faso
• households survey

JEL:

• O13
• O18
• D12

Acknowledgments

We are grateful to the survey support team during the fieldwork in Ouagadougou, particularly Dr. Bitibaly Soumaila, Economist and Lecturer at the University of Ouagadougou, for the advice and guidance about surveying in the context of Burkina Faso, as well as the references to relevant national statistical and food agencies. We are very thankful to the anonymous reviewers and editors for their extremely thoughtful and detailed comments that contributed to improve the paper.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/13600818.2023.2183943

Notes

1. The Food Vulnerability Survey in Urban Settings (VAMU) is one of those attempts of food security measurements.

2. The DHS contains limited information that allow the computation of households’ food security indicators and the assessment of behavioural responses to fuel prices. It was hence necessary to implement an original survey.

3. See Appendix, section Sampling and survey data for more details.

4. For the same reasons and also because of the breadth of the questions asked (18 in total) the study favours this approach over the simplified version of FAO-FANTA (Coates et al., 2007) and Ballard et al. (2014).

5. Hamilton et al. (1997) and Bickel et al. (2000) provide a full discussion on the methodology.

6. The general specification of the probability $p_{ij}$ that observation i moves to category j of food security status Y in the Ologit model is given by: $p_{ij}=p\left(Y_i=j\right)=p(k_{j-1}\le Y_i^{\ast }\le k_j=F\left(k_j-\alpha X{ }_i^{\prime }\right)-F\left(k_{j-1}-\alpha X{ }_i^{\prime }\right)$.

where Y* is a latent variable with various threshold points, and the value of the observed variable Y depends on whether Y* crosses the threshold point. F is a cumulative distribution function (cdf), X a set of covariates, ${\text{isin}}_i$the error term is assumed to be normally distributed, and $\alpha$ is a set of parameters. Both Ologit and Oprobit are interpreted in a similar way. The only difference is the expression of F, the cumulative distribution function

7. To fit the IV-Oprobit model, this approach uses the fully observed recursive mixed process proposed by Roodman (2011). Beside dealing with endogeneity issues, approach provides considerable flexibility in the parameterisation and has the advantage of combining different types of models into a simultaneous equation system. The estimation of the uses maximum likelihood approximations. Given a link function, $g\left(w\right)=\tilde{w}$, and ascending cut points, ${\mathrm{c}}_{1,}\dots ,\hspace{0.17em}{\mathrm{c}}_{\mathrm{I}-1}$, the likelihood function, ${\tilde{w}}_j$, of the model that contains the endogenous variable in its covariates takes the form of.

${\text{Di=D}}_{\text{t}}^{\text{+}}\left(\text{i-1}\right){\text{xD}}_{\text{t+1}}\text{-DtN},\text{i=1},\text{2},\text{3},\dots \text{N}{L}_j\left(\alpha ,{\sigma }^2,{\text{c}}_{1,}\dots ,{\text{c}}_{\text{I}-1};{\tilde{w}}_j\text{|}{x}_j\right)=\int {\text{c}}_{\text{i}-1}-xj{\text{c}}_{\text{i}}-xj{f}_{\in }(\in )d\in =\int h^{-1}\left({\tilde{w}}_j\right)f_{\in }(\in )d\in$

Where $f_{\text{isin}}$ is a probability distribution for the error $\text{isin}$, and $h$ an error link function between errors and outcomes.

8. Variables such as gasoline tax and domestic oil first purchasing price have been used as instruments by Liu (2016). These variables cannot be used as instruments in this study given its nature: both gasoline tax and domestic oil first purchasing price would appear constant as the study uses cross-sectional data and a city level survey.